Here we model an intervention to improve food environments and address malnutrition among school-aged children in urban Hanoi. Conceptually the food environment is described in the following figure. The circles with the dotted lines are external or underlying factors.

We express the general concept of the food environment is slightly differently when we describe it in the context of an intervention. In this figure we show that both the garden products and the education program of a school garden are expected to influence the food environment and how this is expected to happen. The garden itself may have a direct effect if it produces fruits and vegetables that are consumed by students. The garden may also effect the food environments of children by influencing their knowledge about and desire for healthy foods, including fruits and vegetables. The education that students receive may change their perception, make them willing to pay, and also increase their access to fresh fruits and vegetables. Several other expected effects are also included in the figure.

Conceptual model of how school gardens in urban Hanoi schools could influence the food environments of children.

Garden model

The gardens will be created in existing land that is available on school grounds. The STEM option gardens will be designed to for teaching about nutrition and STEM (an approach to learning and development that integrates science, technology, engineering and maths) at primary and secondary schools.

The model was developed across several iterative workshops in July 2023. These included decision definition, conceptual model development of the selected decision and an initial programming session with preliminary model results. The resulting model was further developed through August to November 2023.

Conceptual model of school gardens as an intervention. Should urban Hanoi school boards invest time and money in creating school gardens? Should they invest in formal STEM education as part of these gardens?

No garden

Here we model the baseline of not having a garden. There are some advantages to this including additional play area for kids and the possibilility of other uses for the space such as parking when school is not in session.

School garden

This intervention involves establishing a garden with a dual focus on gardening education. The initiative encompasses the financial aspects of compost, livestock, garden design, equipment, and construction costs. Establishment costs are allocated for compost initiation, worm maintenance, introducing small animals, hiring a planner for garden design, and acquiring necessary equipment, including potential smart systems. The focus extends from enriching soil quality to creating an aesthetically pleasing and functionally equipped garden

School garden designed to contribute to teaching Science Technology Education and Maths (STEM)

In this option we integrate a focus on sciences and STEM education into the garden. Integrating STEM education through the provision of specialized teaching equipment and training for teachers. Establishment costs include science-oriented teaching equipment, with a focus on practical tools like microscopes. Teachers need training for gardening and STEM topics spanning 5 to 7 subjects over 12 days. The intervention aims to not only create a sustainable and interactive garden space but also to enhance educational experiences by incorporating science-oriented learning.

A test run of the intervention will be carried out by the Center for the Development of Organic Agriculture (CODAS) under the Association of Organic Agriculture of Vietnam. In the 2nd year the garden is expected to start running well. The 3rd year is when the STEM education plan will be fully running.

Urban Hanoi school garden

Here is our simulation of the school garden intervention options.

# Source our model
source("CODAS_Garden_Model.R")

# Ensure consistent results with the random number generator
# not for each 'run' of the MC simulation but for 
# consistency each time we call on the simulation 
set.seed(1234) 

garden_simulation_results <- mcSimulation(
  estimate = estimate_read_csv("inputs_school_garden.csv"),
  model_function = school_garden_function,
  numberOfModelRuns = 1000, #run 1000 times
  functionSyntax = "plainNames"
)

Here is a plot of the Net Present Value (i.e. current value of the future benefits) of the three options. "NPV_garden" is the value of the 5 years of the garden intervention."NPV_garden_STEM" is the same garden but with the additional costs and benefits of a full STEM education program, "NPV_no_garden" is the result of 5 years of using the land for something that is not related to the garden, i.e. as a playground or for parking:

source("functions/plot_distributions.R")
plot_distributions(mcSimulation_object = garden_simulation_results, 
                                    vars = c("NPV_garden","NPV_garden_STEM", "NPV_no_garden"),
                                    method = 'hist_simple_overlay', 
                                    base_size = 7, 
                                    x_axis_name = "Comparative NPV outcomes")

Framing the outcomes

The way we present NPV values can influence decision makers - the same information presented in different ways can lead to different decisions. Here we plot the distribution for the decision and frame the projected NPV gain for the decision. These are distributions for the options, with the NPV values of the no garden option subtracted from those for the garden, the decision and garden with STEM, the decision_STEM. We display this as a “Reduction in potential loss” as it is expected to be more attractive to decision makers due to loss aversion, i.e. school boards might put more emphasis on avoiding potential losses than on seeking gains. We can frame our results as a strategy that minimizes losses rather than one that maximizes gains.

plot_distributions(mcSimulation_object = garden_simulation_results, 
                                    vars = c("decision", "decision_STEM"),
                                    method = 'hist_simple_overlay', 
                                    base_size = 7,  
                                    x_axis_name = "Expected gains with schoolgardens (intervention - no garden)")

Summary of results for the decision

Summary

Here we provide a summary of the garden intervention options. We do this with a summary table of the simulation results. We show the percentage of missing values as well as the mean, median and standard deviation (SD) for each output of our model simulations. We use the gt_plt_summary() from {gtExtras} and with options from {svglite}.

# Subset the outputs from the mcSimulation function (y) to summarize only on the variables that we want.
# names(garden_simulation_results$x)
mcSimulation_summary <- data.frame(garden_simulation_results$x[2:61],
 # names(garden_simulation_results$x)
                                 garden_simulation_results$y[1:7])

gt_plt_summary(mcSimulation_summary) 
mcSimulation_summary
1000 rows x 67 cols
Column Plot Overview Missing Mean Median SD
discount_rate 3.49.4 0.0% 6.5 6.5 0.9
CV_value 0.010.54 0.0% 0.3 0.2 0.1
inflation_rate 2.812.1 0.0% 7.5 7.6 1.5
size_of_garden 29122 0.0% 74.8 75.0 15.1
expensive_garden_size 83.291.6 0.0% 87.4 87.4 1.5
cost_increase_expensive_garden_size 0.981.15 0.0% 1.1 1.1 0.0
if_students_like 0.360.95 0.0% 0.6 0.6 0.1
if_parents_like 0.311.00 0.0% 0.7 0.7 0.1
if_community_likes 0.040.98 0.0% 0.5 0.5 0.2
if_effective_manage 0.400.77 0.0% 0.6 0.6 0.1
if_garden_yield_enough 0.170.71 0.0% 0.5 0.5 0.1
if_garden_healthy 0.321.00 0.0% 0.7 0.7 0.1
if_teachers_like 0.020.99 0.0% 0.5 0.5 0.2
if_effective_teaching 0.010.99 0.0% 0.6 0.6 0.2
if_effective_training 0.031.00 0.0% 0.5 0.5 0.2
if_offer_green_space 0.260.99 0.0% 0.7 0.7 0.1
if_reduce_polution 0.090.69 0.0% 0.3 0.3 0.1
if_biophysical_good 0.070.68 0.0% 0.4 0.4 0.1
equipment_cost 28120 0.0% 74.9 74.7 15.5
construction_cost 936 0.0% 22.6 22.7 4.5
garden_designing_costs 7.316.8 0.0% 12.5 12.5 1.5
teacher_training_cost 029 0.0% 12.7 12.6 4.5
school_board_planning 315 0.0% 9.0 9.0 1.8
teaching_equipment 3.112.3 0.0% 7.4 7.4 1.5
compost_starting 2.612.6 0.0% 7.5 7.5 1.6
worm_starting 1.06.1 0.0% 3.5 3.5 0.9
livestock_costs 0.86.6 0.0% 3.5 3.5 0.9
if_family_pays_establishment 0.030.68 0.0% 0.4 0.3 0.1
establishment_family_portion_paid 0.050.63 0.0% 0.3 0.3 0.1
maintaining_labor 1950 0.0% 32.7 32.7 4.5
teacher_salary_cost 1533 0.0% 25.0 25.0 3.1
teaching_equipment_annual 3.413.1 0.0% 7.6 7.6 1.5
teaching_tools 0.66.5 0.0% 3.5 3.5 0.9
seed_costs 0.62.4 0.0% 1.5 1.5 0.3
fertilizer 0.62.3 0.0% 1.5 1.5 0.3
plant_protection 1.06.7 0.0% 3.5 3.5 0.9
livestock_maint 013 0.0% 5.8 5.8 2.4
annual_teacher_training 1.76.1 0.0% 4.0 4.0 0.6
if_school_has_canteen 0.060.64 0.0% 0.4 0.4 0.1
canteen_savings 213 0.0% 7.6 7.6 1.5
sale_of_yield 239 0.0% 20.1 20.4 6.1
extra_cirricular_savings 0144 0.0% 60.1 59.8 24.9
formal_edu_savings 035 0.0% 9.0 8.3 5.8
formal_edu_savings_STEM 2144 0.0% 61.6 62.0 24.4
outside_investment_value 0237 0.0% 34.3 24.0 34.0
outside_investment_value_STEM 01K 0.0% 179.8 130.3 169.2
increased_enrollment_value 034 0.0% 9.1 8.4 5.9
increased_enrollment_value_STEM 0140 0.0% 51.8 50.6 27.2
if_increase_tuition 0.0020.036 0.0% 0.0 0.0 0.0
if_increase_tuition_STEM 0.010.38 0.0% 0.2 0.2 0.1
tuition_increase 2.912.0 0.0% 7.5 7.4 1.5
child_veg_access 3.012.2 0.0% 7.5 7.5 1.5
child_healthier_choices 1141 0.0% 59.1 57.8 24.6
child_healthier_choices_STEM 2734 0.0% 292.9 292.9 118.8
green_space_value 55246 0.0% 149.8 150.3 30.5
reduce_polution_value 725 0.0% 15.1 15.1 3.0
school_event_value 071 0.0% 29.3 28.9 12.3
school_event_freq 015 0.0% 5.8 5.7 2.4
value_of_non_garden_land_use 866 0.0% 35.1 35.2 8.7
if_parking 0.010.20 0.0% 0.1 0.1 0.0
NPV_garden -5972K 0.0% 204.2 155.3 344.9
NPV_garden_STEM -4184K 0.0% 984.0 902.3 623.6
NPV_no_garden 592K 0.0% 344.6 248.0 304.2
decision -1.8K1.2K 0.0% −140.4 −123.2 453.7
decision_STEM -1.3K3.3K 0.0% 639.5 591.5 682.1
total_costs 4231K 0.0% 861.0 856.6 129.3
total_costs_STEM 6332K 0.0% 1,091.5 1,085.8 132.6
# a summary table with missing, mean, median and sd

Summary of the savings for the passive education garden option

summary(garden_simulation_results$y$decision)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## -1833.1  -358.7  -123.2  -140.4   136.3  1171.5

Summary of the savings for the formal STEM education garden option

summary(garden_simulation_results$y$decision_STEM)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## -1252.1   192.2   591.5   639.5  1077.1  3334.1

Summary of costs

summary(garden_simulation_results$y$total_costs)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   422.6   766.1   856.6   861.0   947.3  1297.2
summary(garden_simulation_results$y$total_costs_STEM)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   632.8   999.8  1085.8  1091.5  1183.2  1523.1

First year

summary(garden_simulation_results$y$Cashflow_garden1)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## -258.53 -120.96  -75.61  -69.13  -25.62  234.47
summary(garden_simulation_results$y$Cashflow_garden_STEM1)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## -226.12  -44.62   27.02   44.38  120.22  596.81

Cashflow of the garden option without formal STEM education

source("functions/plot_cashflow.R")
plot_cashflow(mcSimulation_object = garden_simulation_results, 
              cashflow_var_name = "Cashflow_garden")

Cashflow of the garden option with formal STEM education

source("functions/plot_cashflow.R")
plot_cashflow(mcSimulation_object = garden_simulation_results, 
              cashflow_var_name = "Cashflow_garden_STEM")

Expected Value of Perfect Information (EVPI)

Here we assess value of information with the multi_EVPI function.

# Subset the outputs from the mcSimulation function (y) by selecting the correct variables be sure to run the multi_EVPI only on the variables that we want. Find them with names(garden_simulation_results$y)
mcSimulation_table <- data.frame(garden_simulation_results$x, 
                                 garden_simulation_results$y[1:5])

Value of information for the garden option (no STEM).

source("functions/multi_EVPI.R")
# first_out_var is the first result variable in the table, "NPV_garden" in our case.
evpi <- multi_EVPI(mc = mcSimulation_table, first_out_var = "NPV_garden")
## [1] "Processing 5 output variables. This can take some time."
## [1] "Output variable 1 (NPV_garden) completed."
## [1] "Output variable 2 (NPV_garden_STEM) completed."
## [1] "Output variable 3 (NPV_no_garden) completed."
## [1] "Output variable 4 (decision) completed."
## [1] "Output variable 5 (decision_STEM) completed."
source("functions/plot_evpi.R")
plot_evpi(evpi, decision_vars = "decision")

Value of information for the garden option with formal STEM education.

# using the results of the same multi_EVPI
plot_evpi(evpi, decision_vars = "decision_STEM")
## Warning: There are no variables with a positive EVPI. You probably do not need
## a plot for that.

In this case there were no high value variables. We can be onfident in supporting the decision, even with our imperfect current understanding.

PLS

We use Projection to Latent Structures model to get some sense of the correlation strength and direction for model variables and our outcome variables.

For passive education garden option

source("functions/pls_model.R")
pls_result <- pls_model(object = garden_simulation_results,
                                resultName = names(garden_simulation_results$y)[1], # the "NPV_garden" 
                                ncomp = 1)
# read in the common input table
input_table <- read.csv("inputs_school_garden.csv")
# source the plot function
source("functions/plot_pls.R")
plot_pls(pls_result, input_table = input_table, threshold = 0.9)

For school garden with formal STEM education

pls_result_STEM <- pls_model(object = garden_simulation_results,
                                resultName = names(garden_simulation_results$y)[2], # the "NPV_garden_STEM" 
                                ncomp = 1)

plot_pls(pls_result_STEM, input_table = input_table, threshold = 0.9)

The full repository can be accessed at https://github.com/CWWhitney/nifam_codas_school_garden with the following QR code.